Method and System Operating an Imaging System in an Image Capturing Device Based on Artificial Intelligence Techniques

1. A method for image capture, the method comprising: classifying a noise associated with a lens of an imaging device during image capture; preprocessing a captured image based on the classified noise to determine an initial region of interest (ROI) with respect to the captured image, the preprocessing comprising determining a negative-image of the captured image and applying at least one noise filter to the negative-image, the at least one noise filter corresponding to the classified noise; generating a first processed image by inputting the initial ROI and the captured image to a deep learning network, the first processed image corresponding to a first reconstruction of the captured image; rendering a first preview corresponding to the first processed image; receiving a passive user input corresponding to a portion of the first preview; determining a second ROI with respect to the captured image based on the passive user input and the classified noise, the second ROI positioned at least partially within the initial ROI; generating a second processed image by inputting the second ROI and the captured image to the deep learning network, the second processed image corresponding to a second reconstruction of the captured image; and generating a user-personalization based noise-corrected final image based on the second processed image.

2. The method of claim 1, wherein the classifying the noise associated with the lens of the imaging device comprises: capturing a live-view image as the captured image; calculating at least one metric based on the live-view image; receiving a notification that indicates a visual artifact in the live-view image based on the calculated at least one metric; determining if a lens of the imaging device is externally contaminated based on the notification; and classifying the external lens contamination as one or more type of noise through a convolution neural network (CNN) based noise classifier.

3. The method of claim 2, wherein preprocessing of the captured image comprises: computing the negative-image based on a color space to highlight a noise affected region in the captured image; applying one or more noise localization mechanism to the negative-image based on selection of the at least one noise filter in turn based on the classified noise; and determining the initial ROI for the captured images-image based on the applying of the applied one or more noise localization mechanisms.

4. The method of claim 3, wherein the at least one noise filter applied to the negative-image comprises at least one of: a fog filter comprising a linear transformation criteria based at least in part on a brightness, contrast and luminance associated with the negative-image; and a water or dust filter comprising an edge detection criteria based on one or more medians of the negative-image, a morphological closing process comprising a plurality of shape expansion operations and shape reduction operations, and outlier-removal.

5. The method of claim 1, wherein the preprocessing the captured image comprises: extracting, via a plurality of hardware parameter prediction blocks, at least one vector for the captured image based on the classified noise, the at least one vector comprising at least one of a significant component vector and a hardware parameter vector, and the plurality of hardware parameter prediction blocks corresponding to pre-trained networks configured to correct an image affected by at least one of dust, fog, and water droplets; and predicting, via the plurality of hardware parameter prediction blocks, a first plurality of commensurate hardware parameters based on the at least one vector.

6. The method of claim 5, wherein the rendering the first preview comprises: processing the first processed image based on the predicted first plurality of commensurate hardware parameters.

7. The method of claim 1, wherein the determining the second ROI with respect to the captured image comprises: determining location coordinates with respect to the first processed image, based on the portion of the first preview corresponding to the passive user input, by sensing at least one of a contact with a region of a display on which the first preview is displayed, the contacted region of the display being proximate to the portion of the first preview, a gesture corresponding to the portion of the first preview, and an utterance corresponding to the portion of the first preview.

8. The method of claim 7, further comprising: based on ascertaining that the location coordinates correspond to a location inside the initial ROI, applying a shape-reduction based operation against the location coordinates to determine the second ROI with respect to the captured image, the second ROI corresponding to a smaller area than the initial ROI.

9. The method of claim 8, further comprising: based on ascertaining that the location coordinates correspond to a location outside the initial ROI, applying a shape-expansion based operation in accordance with the classified noise against the location coordinates to determine the second ROI with respect to the captured image.

10. The method of claim 5, further comprising: modifying the at least one vector based on location coordinates of the portion of the first preview; predicting, via the plurality of hardware parameter prediction blocks, a second plurality of commensurate hardware parameters based on the modified at least one vector; and rendering a second preview corresponding to the second processed image based on the predicted second plurality of commensurate hardware parameters.

11. The method of claim 10, wherein the rendering the second preview comprises: applying the predicted second plurality of commensurate hardware parameters to the second processed image; and generating the user-personalization based noise-corrected final image, based on the processed second processed image.

12. The method of claim 1, wherein the passive user input corresponding to the portion of the first preview comprises at least one of a touch for focus setting and eye gaze information used to determine the second ROI with respect to the captured image.

13. The method of claim 1, wherein the preprocessing of the captured image and the generating of the first processed image correspond to a predetermined criteria.

14. The method of claim 1, wherein the preprocessing of the captured image and the generating of the first processed image correspond to a dynamically determined criteria, the dynamically determined criteria comprising a real-time determination of at least one of a type of a noise-filter for localizing the classified noise, a type of deep learning network, and a type of a hardware parameter applied to the first processed image.

15. An image-capturing device comprising: a noise-classifier module configured to classify a noise associated with a lens of an imaging device during image capture; a first image processing module configured to preprocess a captured image based on the classified noise to determine an initial region of interest (ROI) with respect to the captured image based on a negative-image of the captured image and at least one noise filter applied to the negative-image, the at least one noise filter corresponding to the classified noise; a first image reconstruction module configured to at least partially reconstruct the captured image by generating a first processed image based on inputting the initial ROI and the captured image to a deep learning network; a rendering module configured to render a first preview corresponding to the first processed image; a user feedback module configured to receive a passive user input corresponding to a portion of the first preview; a second image processing module configured to determine an additional a second ROI with respect to the captured image based on the passive user input and the classified noise, the second ROI positioned at least partially within the initial ROI; a second image reconstruction module configured to reconstruct the captured image by generating a second processed image based on inputting the second ROI and the captured image to the deep learning network; and an output generating module configured to generate a user-personalized and noise-corrected final image.

16. The image-capturing device of claim 15, further comprising a hardware correction module configured to: extract at least one vector for the captured image based on the classified noise, the at least one vector comprising at least one of a significant component vector and a hardware parameter vector; and predict a first plurality of commensurate hardware parameters based on the at least one vector.

17. The image-capturing device of claim 16, wherein the hardware correction module is further configured to: modify the at least one vector based on location coordinates of the portion of the first preview; and predict a second plurality of commensurate hardware parameters based on the modified at least one vector.

18. The image-capturing device of claim 17, wherein the rendering module is further configured to: apply the predicted second plurality of commensurate hardware parameters to the second processed image to render a second preview corresponding to the second processed image; and generate the user-personalization based noise-corrected final image based on the processed second processed image.

19. The image-capturing device of claim 15, wherein the first image processing module is configured to apply the at least one noise filter to the negative-image based on a dynamically determined criteria comprising a real-time determination of at least one of a type of a noise-filter for localizing the classified noise, a type of deep learning network, and a type of a hardware parameter applied to the first processed image.

20. A non-transitory computer readable medium for storing computer readable program code or instructions which are executable by a processor to perform a method for image capture, the method comprising: classifying a noise associated with a lens of an imaging device during image capture; preprocessing a captured image based on the classified noise to determine an initial region of interest (ROI) with respect to the captured image based on a negative-image of the captured image and at least one noise filter applied to the negative-image, the at least one noise filter based on the classified noise; generating a first processed image by inputting the initial ROI and the captured image to a deep learning network, the first processed image corresponding to a first reconstruction of the captured image; rendering a first preview corresponding to the first processed image; receiving a passive user input corresponding to a portion of the first preview; determining a second ROI with respect to the captured image based on the passive user input and the classified noise, the second ROI positioned at least partially within the initial ROI; generating a second processed image by inputting the second ROI and the captured image to the deep learning network, the second processed image corresponding to a second reconstruction of the captured image; and generating a user-personalization based noise-corrected final image based on the second processed image.

21. The method of claim 1, wherein the initial ROI is generated based on the at least one noise filter applied to the negative-image.

22. The method of claim 1, wherein the first processed image is generated by obtaining a first mask based on the initial ROI.

23. The method of claim 22, wherein the second processed image is generated by obtaining a second mask based on the second ROI.